Applications Performance on Reconfigurable Computers by

Field-programmable gate arrays have made reconfigurable computing a possibility. By placing more responsibility on the compiler, custom hardware constructs can be created to expedite the execution of an application and exploit parallelism in computation. Many reconfigurable computers have been built to show the feasibility of such machines, but applications, and thus compilers required to map those applications to the hardware, have been lacking. RAWCS (Reconfigurable Architecture Workstation Computation Structures) is a methodology for implementing applications on a generic reconfigurable platform.[3] Using RAWCS, two applications, sorting and the travelling salesperson problem, are studied in detail. The hardware constructs and techniques required to map these applications to a reconfigurable computer are presented. These techniques involve the numerical approximation of complex mathematical operations in the simulated annealing solution to the travelling salesperson problem in order to make it possible to map those operations to a reconfigurable platform. It is shown that these integer approximations have no adverse effects on the solution quality or computation time ove